Quantitative Partial Acquisition Slice Alignment

QPASA is a graphical tool allowing near identical slice placement across individuals despite differing morphology. The tool walks through (1) coregistation of an outlined oblique slice (in standard MNI space) to subject space in real time and (2) uploading the subject specific image to the scan computer to guide slice placement.

It is written in python/tk and guides the user through brain extraction (skull stripping) and linear warping an atlas to native space both via FSL tools. A Matlab script is used to write the generated subject atlas as dicom files that can be imported back to the scanner for guided slice placement. AFNI's gui is used for interactively exploring the slice warpped from standard to subject space.

Install

The quickest way to get the qpasa command is to run

pip install git+https://github.com/LabNeuroCogDevel/QPASA_slice-warp.git

(and have e.g. ~/.local/bin in your $PATH)

Alternatively, you can clone and run from the repo directory like

git clone https://github.com/LabNeuroCogDevel/QPASA_slice-warp.git
cd QPASA_slice-warp.git
./qpasa

Depends

• python3: nipy, pydicom, numpy, nibabel
• fsl (skullstrip, flirt)
• afni (for viewing final nifti)
• matlab (for rewriting nifti to dicom - python implemetation not accepted by siemens scanner?)
• optionally ROBEX (slow but accurate skull stripping)
• scanner exported samba share mounted to exchange dicom directories
• a way to launch python scripts (windows, see anaconda; osx, consider automator running homebrew python)

Tutorial

The goal is to

1. read in t1 dicoms
2. put a plane through the anatomical to indicate where to postion a slice
3. make new dicoms with the annotation avaiable to the scanner

We can guide the GUI to do this:

Open

When opening slice_warp.py, you are prompted for an example anatomical dicom.

• All dicoms in this folder will be used to reconstruct a nifti file.
• N.B. The final step will save dicoms to a new folder as a sibling to the one you selected. The input and output directory will share the parent directory highlighted in purple here.

Skullstrip

After selecting a reprsentive dicom,

• an 3d image will be constructed and skull stripped with the bet default fractional intensity threshold of .5.

The default skull stripping can be improved by moving the slider and clicking 're-strip'.

smaller values give larger brain outline estimates

Warp

Once we have the brain isolated from skull, we are read to warp. Clicking this button will

• bring our slice mask into native space
• change the brain image area to display the annotation plane mask in red
• Warping takes 30 seconds.

Make

After warping, we need to make a new image that includes the slice and convert it back to dicom for the scanner. After pushing make,

• matlab is launched to convert annotated nifti back to dicom
• we can inspect the newly created image with AFNI. This is also useful for setting the slice position by hand if the following 'copy back' step does not work.

Copy

Finally, we can copy the newly created dicoms back to the scanner, assuming the scanner can read from the parent of the directory of the initial dicoms (highlighted purple in the first image).

• The new output folder will have _mlBrainStrip_ in the directory name
• dicoms will have a sequence number 200 greater than the initial input dicoms.

Files

• warps.bash and the incluced nifti images (*.nii*) follow the FSL pipeline to accomplish this.
• slice_warp.py is a guided GUI to select a dicom folder, bet, and warp.
• rewritedcm.m and rewritedcm.py are attempts to write the output nii back to dcm. see rewritedcm.bash for usage. both are also called by slice_warp.py
• mkOSxApp.bash failed attempt to create a launcher for python "app"

Testing

python3 -m doctest *.py 
python3 -m pytest test/
python3 -m pytest test/test_pipeline.py -k test_matlab --pdb